As the demand for AI networking solutions continues to grow, driven by the bandwidth-intensive GPUs needed to support artificial intelligence technologies, businesses are reevaluating their network infrastructures.
Juniper Networks Inc., which was recently acquired by Hewlett Packard Enterprise, is focused on helping customers construct their own network infrastructures.
Juniper Networks showcases how it is harnessing AI at the New York Stock Exchange.
“We are dealing with the … customers who already know that their data privacy is a big issue,” said Praveen Jain (pictured), senior vice president and general manager of AI clusters and cloud-ready data center at Juniper. “They are already interested in building their own on-prem data centers because of data privacy or any other reasons they may have. These are the customers who are in more advanced stages. They know that they have to build it, but even those customers are the customers who need help because AI is so new. Nobody knows how to operate this infrastructure. What is the right infrastructure, what is the performance, what’s the cost it’s going to bring to the table because GPUs are so expensive?”
Jain spoke with theCUBE Research’s Bob Laliberte at the Juniper Networks AI-Native NOW event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed AI networking solutions and how Juniper is meeting the demands of the AI era. (* Disclosure below.)
AI networking solutions: Simplifying with network assistant Marvis
Deploying AI across a network is difficult because it requires far more bandwidth and can cause congestion. To counteract these issues, Juniper uses the Marvis Virtual Network Assistant, harnessing Mist AI to automatically identify the cause of network failures and simplify its operations.
“What we did differently than any networking vendor in the industry is we built our own network,” Jain said. “I’m the guy who wants to [get] our hands dirty. We’re not just talking. We built our own network. In that network, we brought in Nvidia GPUs. We brought in ecosystem partners like Weka Storage. We brought in compute, we brought in other elements all together. Not only did we build our network, but also we used that network to train the workloads.”
Juniper allows companies to bring in their own data and create networks with the Juniper Validated Design approach, based on its Ethernet ecosystem. Fifty-nine percent of companies prefer Ethernet, according to findings by theCUBE Research, and Jain emphasizes that it is the future of AI networking.
“At the end of the day, it’s pure Ethernet,” he said. “We are adding some Ultra Ethernet capabilities into this new network … it’s easy to operate that. We have Mist Marvis technology at the core, which we applied to our regular data centers. Now we are applying it to a networking for AI solution. That’s where the entire solution is catered towards simplification.”
Here’s the complete video interview, part of SiliconANGLE’s and theCUBE Research’s coverage of the Juniper Networks AI-Native NOW event:
(* Disclosure: TheCUBE is a paid media partner for the Juniper Networks AI-Native NOW event. Neither Juniper Networks Inc., the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)